Update dataset card

README.md

@@ -10,42 +10,22 @@ configs:
 - config_name: default
   data_files:
   - split: train
-    path: data/train-*
+    path: data/train-*.parquet
 tags:
 - simple-wikipedia
 - wikipedia
 - markdown
 - sqlite
-dataset_info:
-  features:
-  - name: page_id
-    dtype: int64
-  - name: title
-    dtype: string
-  - name: content
-    dtype: string
-  - name: content_no_link
-    dtype: string
-  - name: importance
-    dtype: string
-  - name: truncated
-    dtype: bool
-  - name: error
-    dtype: bool
-  splits:
-  - name: train
-    num_bytes: 6993133
-    num_examples: 1000
-  download_size: 4042374
-  dataset_size: 6993133
 ---
 
 # Simple English Wikipedia (Markdown)
 
+Recurring weekly snapshot of Simple English Wikipedia (https://simple.wikipedia.org/), which uses shorter sentences and limited vocabulary compared to the main English Wikipedia. This makes it smaller, easier to parse, and better suited for on-device or bandwidth‑constrained assistants while still covering broad general knowledge. Ideal as an offline Wikipedia MCP server backing a household AI assistant.
+
 - Dump date: 2025-12-01
 - Source dump: https://dumps.wikimedia.org/simplewiki/20251201/simplewiki-20251201-pages-articles.xml.bz2
 - SHA-1: ee583946e86857e9f1e155f80bd3cd8b5d6dade7
-- Records: 276856
+- Records: 1000
 - Refresh cadence: Weekly on Sundays at 11:00 UTC
 
 ## Dataset Structure
@@ -62,7 +42,7 @@ Columns:
 ## Processing
 
 - Downloaded `pages-articles` XML dump and verified SHA-1.
-- Kept only namespace 0 articles and skipped redirects.
+- Kept only namespace 0 articles, skipped redirects, and dropped titles beginning with “List of”.
 - Stripped templates/ref/gallery blocks and file/category links; converted headings, lists, tables, and internal/external links to Markdown with page IDs.
 - Stored a SQLite mirror (`pages` table) alongside the Hugging Face dataset.
 - Markdown links point to the target page's numeric ID for fast lookup without a title-to-ID join.
@@ -78,4 +58,28 @@ ds = load_dataset("juno-labs/simple_wikipedia", split="train")
 print(ds[0])
 ```
 
-The SQLite file `simplewiki-20251201-articles.sqlite` contains identical columns for tooling that prefers SQL.
+SQLite usage (`simplewiki.sqlite` mirrors the same columns):
+
+```bash
+sqlite3 simplewiki.sqlite "SELECT page_id, title, substr(content,1,200) || '...' FROM pages LIMIT 5;"
+```
+
+You can also mount it in code:
+
+```python
+import sqlite3
+conn = sqlite3.connect("simplewiki.sqlite")
+cur = conn.cursor()
+for row in cur.execute("SELECT title, content FROM pages WHERE page_id = ?", (7553,)):
+    print(row)
+```
+
+
+## Categorization
+
+Model: openai/gpt-5-mini
+Prompt template:
+```
+You are an expert product manager for a household smart-speaker AI in the United States. Classify how important (to one of: "low", "medium", or "high") it is to store this article offline for day-to-day user queries. Respond with JSON containing only a "label" field set to one of: "low", "medium", "high".
+```
+